VisionLanguageModelOptions
Options & Configuration · AiDotNet.NeuralNetworks.Options
Shared native-tower configuration for vision-language and multimodal models (BLIP, BLIP-2, Flamingo, LLaVA, ImageBind, GPT-4 Vision, VideoCLIP and relatives).
Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → VisionLanguageInputOptions → VisionLanguageModelOptions
Known derived types: AudioVisualCorrespondenceOptions, AudioVisualEventLocalizationOptions, Blip2Options, BlipOptions, FlamingoOptions, Gpt4VisionOptions, ImageBindOptions, LLaVAOptions, UnifiedMultimodalNetworkOptions, VideoCLIPOptions
For Beginners
These models read pictures and text together. The settings here describe how big the pictures are, how they get chopped into patches, how much text the model can read at once, and how wide its internal representations are. Each model’s own options class supplies its shipped defaults.
How It Works
Derived options classes assign their shipped values in their parameterless constructor.
See ModelHyperparameterOptions for why these properties are non-nullable.
ONNX-only input contracts use VisionLanguageInputOptions instead. Native
tower settings do not resize a loaded ONNX graph.
Properties
| Property | Summary |
|---|---|
int Channels { get; set; } |
Gets or sets the number of colour channels in the input image. |
int HiddenDim { get; set; } |
Gets or sets the width of the model’s main hidden representation. |
int NumEncoderLayers { get; set; } |
Gets or sets the number of layers in the text encoder. |
int NumHeads { get; set; } |
Gets or sets the number of attention heads. |
int PatchSize { get; set; } |
Gets or sets the side length, in pixels, of each square patch the image is divided into. |
int VisionDim { get; set; } |
Gets or sets the width of the vision tower’s hidden representation, which is often wider than the shared embedding space. |
int VisionLayers { get; set; } |
Gets or sets the number of layers in the vision tower. |
int VocabSize { get; set; } |
Gets or sets the number of distinct tokens the text tokenizer covers. |